BepFAMN: A Method for Linear B-Cell Epitope Predictions Based on Fuzzy-ARTMAP Artificial Neural Network.

Area of Science:

• Immunoinformatics
• Computational Biology
• Machine Learning in Immunology

Background:

• Public health relies heavily on vaccines, necessitating accurate diagnostic methods for disease monitoring and vaccine development.
• Experimental epitope mapping is laborious, time-consuming, and costly, especially for large pathogens.
• Existing in silico methods for epitope prediction suffer from low accuracy and slow, offline training.

Purpose of the Study:

• To develop an accurate and efficient computational method for predicting B-cell linear epitopes.
• To improve upon existing machine learning-based epitope prediction tools.
• To enable online training for continuous model improvement and reduced experimental validation.

Main Methods:

• Developed BepFAMN, a Fuzzy-ARTMAP neural network architecture for B-cell linear epitope prediction.
• Trained BepFAMN using 15 amino acid properties, including physicochemical scales, on data from the IEDB.
• Employed five-fold cross-validation and an independent test set (BepiPred-2.0) for rigorous evaluation.

Main Results:

• BepFAMN achieved high performance on the validation dataset: sensitivity=91.50%, specificity=91.49%, accuracy=91.49%, MCC=0.83, AUC ROC=0.9289.
• The testing dataset showed significant results: sensitivity=81.87%, specificity=74.75%, accuracy=78.27%, MCC=0.56, AUC ROC=0.7831.
• BepFAMN outperformed existing linear B-cell epitope prediction tools and supports online training.

Conclusions:

• BepFAMN represents a significant advancement in B-cell linear epitope prediction accuracy and efficiency.
• The online training capability allows for dynamic model updates, reducing the need for complete retraining.
• This method accelerates the identification of epitopes for diagnostic tests, vaccines, and immunotherapeutics, saving time and resources.